Recently, trials of trastuzumab in the adjuvant setting have been completed in both the US (NCCTG/NSABP) and in Europe (the HERA trial) that have shown this drug can decrease the recurrence rate in patients with HER2 positive breast cancer. However in both trials, there are still a significant percentage of patients that recur on the drug. Similarly, numerous earlier trials in patients with metastatic cancer also showed a lack of response even though they had HER2 expressing tumors. These facts, combined with the facts that new drugs are now available that target HER family signaling pathways, suggest that new, more specific, companion diagnostics could be developed for trastuzumab that increase the specificity of selection of patients for this therapy. The underlying hypothesis of this proposal is that an optimal predictor of outcome for patients on trastuzumab can be achieved by combining multiple markers which predict response. We propose that using a set of novel techniques to interrogate HER2 tumors being treated with trastuzumab and chemotherapy, we can define the optimal predictor and validate this classifier in a prospective trial. The techniques will include assessment of DNA, RNA and protein to determine the best modality for optimal prediction. Protein will be assessed using multiplexed immunofluoresence, RNA will be assessed using transcriptional microarray profiling and DNA will be assessed in the form of copy number analysis using high density SNP arrays. Each of these assays has the potential to be translated into a usable companion diagnostic assay for breast cancer patients. In this revised 2 year version. of this grant we propose to keep the analysis of all 3 modalities, but to only complete the training set aspects of the grant. We envision a follow-up submission in 18-24 months that shows the model resulting for the efforts of this project, then proposing validation of the model(s) on the CALGB 40601 cohort or similar. The scaled aims include: AIM 1) To develop an optimal multiplexed predictive model that uses HER pathway related proteins, downstream signaling proteins and hetero- or homodimerization state of HER2. This aim will use the quantitative multiplexing technology (called AQUA) for accurate in situ measurement of protein expression on a series of 10-25 HER2 pathway related proteins to construct a series of models that predict response to trastuzumab in the CALGB 9840 trial (A trial of taxol and trastuzumab in in the first line metastatic setting) AIM 2) To use Ilumina DASL-based Gene Expression Profiling to develop an optimal multiplexed predictive model. This aim will assess gene expression using a custom gene set on the Illumina DASL custom array (1536 genes) Platform in the CALGB 9840 cohort to identify candidate predictors for the multiplexed predictive model. Specifically, candidate amplicons, and genes associated with these candidate amplicons, which are associated with response to trastuzumab will be identified for inclusion in the model. AIM 3) To do computational modeling of the combined data from Aim I and 2 to discover the best 3-5 models that fit the training set (CALGB 9840) data. This aim will create a series of optimal models that best select responders from non-responders in the CALGB training set. Validation of the models will be done by Leave One Out Cross Validation methods in anticipation of future more robust validation in an independent cohort in a subsequent study.